Design Of Multi-Biometrics System Based On Face And Ear

With the recent rapid development, biometrics technologies have already been applied to many areas of social life, such as public security, information security, electronic transactions fields and so on. Nevertheless, the systems use a single biometric have to contend with a variety of problems, such as noisy data, non-universality and spoof attacks. Multi-biometrics systems seek to alleviate and even overcome some of these drawbacks by integrating multiple biometrics from the same subject. Further, there will be a significant increase in recognition rate while the Multi-biometrics systems using an effective data fusion scheme to combine the different biometrics.Face recognition and ear recognition are kind of non-intrusive biometrics, which do not need active cooperation of subject, and the data can be captured at a distance. These advantages make the systems become more acceptable. At present, auto-detection of human face is still a very challenging issue, and even the human ear automatic detection is very little coverage. Under these circumstances, most of researches on face and ear based muliti-biometrics can only rely on manual position and segmentation. This paper will dedicate to the research of automatic detection and localization of human face and ear, and design a fully automated non-intrusive multi-biometrics system based on face and ear.A method for fast automatic eye localization is introduced in this paper. It applies a cascade classifier based on AdaBoost to detect human face and defines the two rough eye regions at first step, and then separates possible eyebrows from the regions with the help of some improved projection function. In the end, the eyes positions would be defined after using a special scheme based on the mean, complexity and structure central measurement information. Also, an automatic ear detection method based on improved GVF Snake algorithm is proposed. Using the horizontal component and vertical component of GVF to build up an ear-shaped map, from which it is easy to extract an initial contour of high quality for GVF Snake to fitting the real ear outer profile. This method achieves a high detection rate and better robustness in experiments, and is proved to be an effective approach to locate ears. Finally, a human face and ear based multi-biometrics system is achieved in VC 6.0 software environment. The system employs the detection methods mentioned above and uses PCA and KPCA to extract face and ear features. Besides two independent single biometrics modules of face and ear, it fulfills several recognition approaches with different fusion schemes in serial and parallel constructions. Meanwhile, it is able to operate in one of three modes: verification, identification (close set mode and open set mode). Overall, the system is a versatile and practical non-intrusive multi-biometrics system with high degree of automation.